bart-base-multi-news

This model is a fine-tuned version of facebook/bart-base on the multi_news dataset. It achieves the following results on the evaluation set:

• Loss: 2.4147
• Rouge1: 26.31
• Rouge2: 9.6
• Rougel: 20.87
• Rougelsum: 21.54

Intended uses & limitations

The inteded use of this model is text summarization. The model requires additional training in order to perform better in the task of summarization.

Training and evaluation data

The training data were 10000 samples from the multi-news training dataset and the evaluation data were 500 samples from the multi-news evaluation dataset

Training procedure

For the training procedure the Seq2SeqTrainer class was used from the transformers library.

Training hyperparameters

The Hyperparameters were passed to the Seq2SeqTrainingArguments class from the transformers library.

The following hyperparameters were used during training:

• learning_rate: 5.6e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum
2.4041	1.0	1250	2.4147	26.31	9.6	20.87	21.54

Framework versions

• Transformers 4.30.0
• Pytorch 2.0.1+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3